[IronPython] static source2handler mapping causing memory leaks

[Dino Viehland]

If I remember correctly we can't just directly hook the events the 'normal' way because then the unhook fails in certain cases.  But it looks like we still have some bugs related to that anyway.  I tracked down the original bug and the way this came about was we had code like:

>>> class C:
...     def cf(self, x):
...             print "inside C.F"
...             return x
...
>>> r = IronPythonTest.ReturnTypes()
>>> c = C()
>>> r.floatEvent += c.cf
>>> r.RunFloat(20)
inside C.F
20.0
>>> r.floatEvent -= c.cf
>>> r.RunFloat(20)
inside C.F
20.0

And before 1.0 beta 3 this was broken.  I fixed it by using the WeakHash class. The reason why the above code was broken is because the CLR only does object equality comparisons when unhooking delegates.  And each time we have a delegate to 'c.cf' we have a new ReflectedMethod which is bound to the instance - and therefore the instances don't compare as equal and the CLR doesn't think the delegates are the same.  Unfortunately we still seem to have some other bugs on CodePlex related to event hooking so I'm guessing this gets another pass before 2.0 is done.

I've gone ahead and opened bug #14454 to track it http://www.codeplex.com/IronPython/WorkItem/View.aspx?WorkItemId=14454.

Also, when it comes to WeakHash we actually are aware of this problem in general although we hadn't yet seen it creep into IronPython.  We knew it'd be an issue for IronRuby where we have to use the weak hash table much more because you can add to arbitrary objects.  We're trying to get some support from the CLR to make this work correctly.  So solving the general problem might take a while but hopefully we'll get a chance to at least revisit the event handling code.

Thanks for the bug report!

From: users-bounces at lists.ironpython.com [mailto:users-bounces at lists.ironpython.com] On Behalf Of Kamil Dworakowski
Sent: Tuesday, December 18, 2007 6:26 AM
To: users at lists.ironpython.com
Subject: Re: [IronPython] static source2handler mapping causing memory leaks

Why not make values of the mapping weak references too?

Weak ref for value is bad because that would allow the handler to be garbage collected even though the sender is alive. That problem is harder than I thought initially.

Why not store the subscribers on a list attached to the source object?

--
Kamil Dworakowski


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/ironpython-users/attachments/20071218/25948bac/attachment.html>